This report is generated from several Rmarkdown documents where each chapter is a separate markdown report. The folowing are the Rmd files used in analyzing the data.
Transit ridership and highway volumes for the Corridor studies are obtained by running various alternative scenarios in STOPS and SERPM models. Both of these models are developed differently and uses different input data sets. However, some of the data between these models come from MPO and transit agencies. The transit data between STOPS and SERPM is in two different formats. STOPS transit inputs are in GTFS format whereas SERPM inputs are in CUBE PT format. Although these two are not comparable formats, they both should represent transit networks. The other input file is socio-demographic or landuse file. The landuse file exists in both model but again in different format and at different geographic levels. Since the transit ridership estimates for the Corridor studies are expected to reply upon STOPS model, transit networks between STOPS and SERPM are not compared here. However, the landuse data is studied in great detail.
The 2015 socio-economic data is developed through linear interpolation. Currently, the 2015 SE data exists in two different models:
The STOPS SE data is at TAZ level where as SERPM model data is at MAZ level. In order to perform a comparative analysis between these two input data, SERPM MAZ level data is aggregated to TAZ. This document shows all findings at TAZ level. Technically SE data in the two models should be same and should originate from the same source. Since both models (STOPS and SERPM) were constantly being updated the model data for 2040 could be different. This document summarizes those differences.
The 2040 SE data exists in the following model locations:
#path <- getwd()
path <-"/Volumes/C/projects/SERPM_Compare/check_seData"
# data directories
dir <- "Corradino_SEData"
fdot.dir <- "FDOT_June_30_2016"
stops.dir <- "STOPS_SEData"
# file names
# TODO (ans): Replace maz_data.csv files with model_data.csv (which is more comprehensive data)
maz.data.files <- c("2010_maz_data.csv", "2015_maz_data.csv", "2040_maz_data.csv")
fdot.maz.files <- c("maz_data_IN_2040R.csv", "maz_data_IN_2040T.csv")
stops_mpo_shapeFile <- "simplified_MPOTAZPopEmp.shp"
taz_county_file <- "taz_county.csv"
# list of TAZs to check
check_taz <- c(76, 387, 979, 1596, 1598, 2253)
# Save R Objects for later use
save.RData.outputs <- TRUEThe SERPM maz data files are developed by FDOT with feedback from various agencies, including the three MPOs in the region. The future year maz data file is constantly updated to reflect revised population and employment projections. Due to this continuous update there are several versions of 2040 data with significant difference across population, households and employment variables. As a part of Corridor studies effort, it is required to document the source of model data being used as well as validate data.
The model data files delivered by Corradino were reviewed for data consistency across the three horizon years: 2010, 2015 and 2040. Some of the data fields are not consistent across all years. Two fields geoSRate and geoSRateNm exist in some maz_data.csv files but not in all.
The 2015 SE data is developed through linear interpolation and thus the growth rate should always be linear and between 2010 and 2040. This section of the code checks if there are any households in 2015 that drop from 2010 but gain back in 2040 (checks growth rate for linearity). The following table shows households across 2010, 2015 and 2040 (data with 5 hhs difference is ignored).
** The household variable in this maz_data.csv file computed by aggregating PopSyn-3 outputs and thus there is some degree of over/under estimation of households at MAZ level when compared to PopSyn-3 inputs ** The differences shown in the data below are within reasonable range.
check_hh <- data_all_years %>%
mutate_each(funs(replace(.,is.na(.),0))) %>%
mutate(diff_1015 = hh_2015 - hh_2010,
diff_1540 = hh_2040 - hh_2015,
check = ifelse((diff_1015 > 0 && diff_1540 < 0) || (diff_1015 < 0 && diff_1540 > 0), 1, 0)) %>%
filter(check == 1 , abs(diff_1540) > 5, abs(diff_1015) > 5)
check_hh <- check_hh %>%
select(TAZ, hh_2010, hh_2015, hh_2040, diff_1015, diff_1540)
# kable(check_hh, caption = "Zones with Inconsistent Households Trends", digits = 0, format.args = list(big.mark = ","))
datatable(check_hh, caption = "Zones with Inconsistent Households Trends")# Save R Object file
if (save.RData.outputs) {
save(check_hh, file = "table_check_hh.RData")
}The following table shows data for selected TAZ: 76, 387, 979, 1596, 1598, 2253. These zones were selected based on the past review of 2015 zonal data. The current model data shows consistent growth rate across household, population and employment variables between 2010, 2015 and 2040 years.
sel_data <- data_all_years %>%
filter(TAZ %in% check_taz) %>%
select(TAZ, pop_2010, pop_2015, pop_2040,
emp_total_2010, emp_total_2015, emp_total_2040,
hh_2010, hh_2015, hh_2040)
kable(sel_data, caption = "Selected TAZ from SERPM Data", digits = 0, format.args = list(big.mark = ","))| TAZ | pop_2010 | pop_2015 | pop_2040 | emp_total_2010 | emp_total_2015 | emp_total_2040 | hh_2010 | hh_2015 | hh_2040 |
|---|---|---|---|---|---|---|---|---|---|
| 76 | 953 | 994 | 1,207 | 119 | 123 | 142 | 569 | 597 | 745 |
| 387 | 214 | 217 | 217 | 745 | 730 | 655 | 98 | 98 | 99 |
| 979 | 1,689 | 1,684 | 1,690 | 243 | 244 | 251 | 788 | 787 | 780 |
| 1,596 | 499 | 505 | 499 | 402 | 417 | 493 | 241 | 241 | 244 |
| 1,598 | 307 | 382 | 656 | 27 | 28 | 31 | 165 | 173 | 241 |
| 2,253 | 34 | 80 | 277 | 265 | 267 | 275 | 15 | 33 | 122 |
The figures below show growth rates by county for selected input variables: households (hh), population (pop), total employment (emp_total), college enrollement (college), school enrollement (school).
Miami-Dade County: The growth rate looks ok here. The hh, pop and emp growth rates are at 25 percent, college enrollment grwoth is at 20 percent and school is at 5 percent.
Broward County: The hh growth rate looks ok, pop seems a bit low at 12 percent but employment is projected to grow by 5 percent? Need to double check with FDOT. Same issue with College and School too.
Palm-Beach County: The growth rate looks ok here. The hh, pop and emp growth rates are at 30 percent, college and school at 30 and 15 percent respectively.
The two official 2040 maz_data.csv files are downloaded from FDOT website June 2016 :
These two data sets are compared with the Corradino delivered 2040 data to make sure the model is using the official version and to document the source of model data being used for the corridor studies.
Several tabulations were computed to ensure that the model data being used for the corridor studies is from the official FDOT Cost Feasible scenario. Table below shows a comparison between the 2040 Cost Feasible and Corradino 2040 model data, where there are zero differences. Chart below shows a scatter plot of population variable from the two data sets where there it clearly depicts both data sets being same.
mgra TAZ.Corradino hh.Corradino pop.Corradino emp_total.Corradino TAZ.FDOT hh.FDOT pop.FDOT emp_total.FDOT diff.hh diff.pop diff.emp pop.bin —– ————– ————- ————– ——————– ——— ——– ——— ————— ——– ——— ——— ——–
Well, the 2040 LRTP data is significantly different from 2040 Cost Feasible and thus the Corradino data also differed. There are too many zones to display the difference in tabular form. The plot below shows a scatter plot of population data from the two data sets.
| mgra | TAZ.Corradino | hh.Corradino | pop.Corradino | emp_total.Corradino | TAZ.FDOT | hh.FDOT | pop.FDOT | emp_total.FDOT | diff.hh | diff.pop | diff.emp | pop.bin |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 2901 | 43 | 169 | 0 | 2901 | 43 | 172 | 0 | 0 | -3 | 0 | -1000 to -1 |
| 2 | 2902 | 9 | 23 | 1337 | 2902 | 9 | 23 | 1337 | 0 | 0 | 0 | 0 |
| 3 | 2903 | 497 | 1694 | 379 | 2903 | 497 | 1685 | 379 | 0 | 9 | 0 | 1 to 1000 |
| 4 | 2903 | 273 | 984 | 21 | 2903 | 273 | 1001 | 21 | 0 | -17 | 0 | -1000 to -1 |
| 5 | 2903 | 383 | 1306 | 86 | 2903 | 383 | 1308 | 86 | 0 | -2 | 0 | -1000 to -1 |
| 6 | 2903 | 212 | 861 | 14 | 2903 | 212 | 841 | 14 | 0 | 20 | 0 | 1 to 1000 |
The latest South East Florida Regional STOPS model is downloaded from FDOT page and was reviewed. As a part of the review, model landuse data and observed APC counts were checked. Since the Corridor studies use both SERPM and STOPS models, it is important to check and ensure the input data is consistent between the models. The downloaded SEFL STOPS model consists of 2010, 2015, and 2040 population and employment data at TAZ level. As per SEFL STOPS model documentation, the 2014 data computed by interpolating between 2010 and 2040. STOPS model utilizes only population and employment variables and household variable is not used and thus not provided in the data set.
This data is clearly different from SERPM 2015 MAZ data.
# Read data from stops input
shape <- readOGR(paste0(path,"/",stops.dir,"/",stops_mpo_shapeFile), layer = "simplified_MPOTAZPopEmp", verbose = FALSE)
stops_se <- shape@data
stops_sel_data <- stops_se %>%
filter(TAZ_REG %in% check_taz) %>%
select(TAZ_REG, POP_10, POP_15, POP_40,
TOTE_10, TOTE_15, TOTE_40)
kable(stops_sel_data, caption = "Selected TAZ from STOPS Data", digits = 0, format.args = list(big.mark = ","))| TAZ_REG | POP_10 | POP_15 | POP_40 | TOTE_10 | TOTE_15 | TOTE_40 |
|---|---|---|---|---|---|---|
| 76 | 953 | 994 | 1,202 | 119 | 123 | 142 |
| 387 | 214 | 214 | 216 | 745 | 730 | 655 |
| 979 | 1,689 | 1,689 | 1,691 | 243 | 244 | 251 |
| 1,596 | 499 | 498 | 495 | 402 | 417 | 493 |
| 1,598 | 307 | 364 | 649 | 27 | 28 | 31 |
| 2,253 | 34 | 75 | 279 | 265 | 267 | 275 |
Table below shows the number of TAZ by range of pop and emp difference. If there is no difference then it is not tabulated here.
| bin | diff.pop_15 | diff_emp_15 |
|---|---|---|
| -100 to -50 | 14 | NA |
| -20 to 0 | 1445 | 177 |
| -50 to -20 | 50 | NA |
| -500 to -100 | 21 | NA |
| -5000 to -500 | 5 | NA |
| 0 to 20 | 1496 | 885 |
| 100 to 500 | 27 | 2 |
| 20 to 50 | 88 | 1 |
| 50 to 100 | 44 | NA |
| 500 to 5000 | 2 | 1 |
The plot below shows population difference between the two data sets.
The plot below shows employment difference between the two data sets.
The map below shows population difference for 2010, 2015 and 2040 by TAZ between the two data sets. The population for 2010 between the two models (SERPM and STOPS) is the same data where as the 2040 data is different across most of the TAZs. I guess the 2040 difference trickled down into 2015 when interpolated.
The employment data for 2010 between the two models (SERPM and STOPS) is the same data where as the 2015 data is different across most of the TAZs. About 5 TAZs show varying employment data for year 2040. Table below shows those 5 TAZs.
tabulate_emp_diff_2040 <- df %>%
filter(diff_emp_40 != 0) %>%
select(TAZ, emp_total_2040, TOTE_40, diff_emp_40)
kable(tabulate_emp_diff_2040)| TAZ | emp_total_2040 | TOTE_40 | diff_emp_40 |
|---|---|---|---|
| 825 | 8160 | 250 | 7910 |
| 854 | 89 | 0 | 89 |
| 864 | 3070 | 386 | 2684 |
| 1058 | 3000 | 0 | 3000 |
| 2409 | 1831 | 1678 | 153 |
The map below shows population difference for 2010, 2015 and 2040 by TAZ between the two data sets.
In the current SERPM version there are 4 transit networks:
Note: Cube catalog shows one another scenario: 2040 LRTP (2040T), for which no input files are provided.
CUBE Transit Network are stored in LIN format and contains both route attributes and stop attributes in the same file. In order to use better use this information for comparative analysis as well as display subjective routes, the LIN files need to be converted to spatial data formats such as shape files. Converting this is a four step process:
There are a total of 624 routes across these four networks.
This document compares all transit networks and summarizes the differences. The following are the different tabulations:
There are about 473 routes that exists in all scenarios. Although these routes have common name, some of the route attributes vary in Mode and Headways. Some of the routes also show minor alignment changes and those routes are compared to better understand the alignment changes. Route alignments over 1 mile (100 cube spatial units) are plotted.
Table List of routes by mode shows the number of routes by mode across all alternatives. The common routes between 2010 and 2015 didn’t show any differences in the attributes. Although these routes are listed as common routes, there is a mode code difference for about 12 routes between 2010/2015 and 2040CF/2040TC alternatives. All of the differences are in BRT, Express Bus and Local Bus reports.
List of common routes with mode change between 2010/2015 and 2040CF/2040TC
Map of common routes with mode change between 2010/2015 and 2040CF/2040TC
List of common routes with ALIGNMENT CHANGE change between 2010/2015 and 2040CF/2040TC
Map of common routes with ALIGNMENT CHANGE over 1 mile are plotted below. Four comparison sets are shown below:
List of common routes with HEADWAY_1 differences
Although these routes exists in all scenarios, the differences between any two scenario networks is very vast. There are about 106 common routes with different headways between 2010 and 2015. The plot below shows these routes along with the headways (click on the link to popup the attributes). The headway differences are printed to csv files for future reference.
Similar to 2010 vs 2015, other common routes from other scenarios are also compared. Check / uncheck the map keys to reveal this information.
List of common routes with HEADWAY_2 differences Although these routes exists in all scenarios, the differences between any two scenario networks is very vast. There are about 86 common routes with difference in headways (HEADWAY_2) between 2010 and 2015. The plot below shows these routes along with the headways (click on the link to popup the attributes). The headway differences are printed to csv files for future reference.
Similar to 2010 vs 2015, other common routes from other scenarios are also compared. Check / uncheck the map keys to reveal this information.
List of common routes with HEADWAY_3 differences Although these routes exists in all scenarios, the differences between any two scenario networks is very vast. There are about 127 common routes with difference in headways (HEADWAY_3) between 2010 and 2015. The plot below shows these routes along with the headways (click on the link to popup the attributes). The headway differences are printed to csv files for future reference.
Similar to 2010 vs 2015, other common routes from other scenarios are also compared. Check / uncheck the map keys to reveal this information.
There are about six routes that exist in 2010 and 2040CF that are not in 2015 and 2040TC. The table below shows the list of these routes along with 2010 and 2040 headway and mode. These routes contain identical attributes between 2010 and 2040.
There are about 22 routes that are present in 2015 and 2040 TC but not in 2010 and 2040CF. These routes are added to support the Flagler Corridor. The routes are located in the City of Doral and Downtown area, which are western and eastern ends of the Corridor. The mode and headway between 2015 and 2040TC for these routes are identical.
There are no routes in 2015 that doesn’t exist in any of the scenarios.
This chapter focuses on checking and reporting of socio-demographic and employment data from 2010, 2015 and 2040.
The regional planning models contains valuable data that can be used beyond travel demand forecasting needs. The forecasted landuse and employment are one of the most useful information for general use. However, in SERPM 7.0 this data exists in a few different places with similar names but with significant variations.
SERPM is an activity based model and is one of the most sophisticated models where coordinated daily activity pattern is the epic center of its design, meaning it is designed to simulate varying travel patterns of each individuals of the households by taking their household socio-demographics and personal profiles into account. This model consists of various sequential sub-modules and use variants of the same data sets as required by each sub-models. For this reason the disaggregate household and population variables such as: zero-car autos, zero-worker households, low-income households, population under age < 16 and over 65 plus can only be found in the output files. However, the zonal level household, population variables are user inputs and can be obtained from the inputs location. Although employment is also a user input file, this can only be found under inputs to ctramp folder, which is at a different location than landuse input files.
As it can be seen, extracting useful general purpose information from the model is different and needs thorough understanding of the SERPM model. To address this issue, R-Script were developed to extract the model data.
The SERPM model input and output files are pre-processed to produce model data file (year_model_data.csv) that are referred in this document. These files are generated by processing the following files:
Note: Script GetPopHH.R is run on SERPM model results for 2010, 2015 and 2040 to extract year_model_data.csv files processes
The first two are outputs from synthetic population generator. The household, population variables in the ctramp/maz_data.csv file doesn’t necessarily agree with PopSyn-3/maz_data.csv, although they correspond to the household and population variables. This difference is due to balancing priorities used in PopSyn-3, where some of the variables are given a higher priority to match to control totals (usually at TAZ level). In cases where zonal data is being analyzed, it is better to use household and population data from input to PopSyn-3 instead of ctramp inputs.
As it is extremely difficult to analyze all the model data by TAZs and MAZs, the data is aggregated by larger geographies such as:
For this report, data was aggregated by counties, districts and corridor study area. The data was not tabulated by cities although it can be done. The districts are defined as aggregated zones and some of the data was collected (AirSage OD trip flows) by these districts. For Flagler Corridor these districts are defined independently and so they are different from districts used in other corridor studies. Other corridors use MPO directed districts and all data collected and analysed is done through this district except for Flalger. The data presented here uses Flagler Corridor district system.
Flagler Corridor study area is comprised of four segments, however the segments 3 and 4 fall under the same districts and so only three segements were defined in this analysis. The code chuck below shows the Flagler Districts by segments. The following are the physical extents of the segments:
• Segment 1 – NW/SW 107th Avenue from Flagler Street to NW 12th Street and NW 12th Street from NW 107th Avenue to NW 122nd Avenue. • Segment 2 – Flagler Street and SW 8th Street from SR 826/Palmetto Expressway to SW 147th Avenue via NW/SW 107th Avenue. • Segment 3 – Flagler Street from NW 27th Avenue to SR 826/Palmetto Expressway • Segment 4 – Flagler Street/1st Street from Downtown Transit Center to NW 27th Avenue
TODOs: Replace the summaries to show trends by each county instead of just Miami-Dade.
2010 Miami-Dade County population is reported at 2.4 million, with 867,000 households and over 1.1 million employees as per Census 2010. 2040 projected growth is estimated at 25 and 30 percent over 2010 values for population and employment respectively. Figure 6 shows the socio-demographic growth for Miami-Dade County across different population groups including transit dependent for the three scenario years: 2010, 2015 and 2040. Most All of the demographic variables show a positive increase.
The Federal Transit Administration defines transit dependent persons as those 1) without private transportation (zero-car households), 2) elderly (over age 65), 3) youths (under age 18), and 4) persons below poverty or median income levels defined by the U.S. Census Bureau.
Table shows Miami –Dade County and Flagler Corridor population by age cohorts and households by zero auto and income categories for the three scenario years: 2010, 2015 and 2040. Based on the Census data for 2010, 2015, and 2040, analysis of each of these sectors of the population was analyzed and summarized in the following sections.
Table below shows the household, demographics summary for the Corridor
Zero car households: This is the sector of the population that has no or limited access to personal vehicles and therefore, are most likely to use transit as a mode of transportation. In 2015, the share of zero car households is 14 percent in the County and 22 percent in the Corridor. Between 2010 and 2015 the zero car households grew by 5,000 and is projected to double by 2040. The growth in zero car households is slightly significantly more by 5% in the Corridor, which is estimated to grow to 30 percent by 2040 compared to County wide growth at 18 percent. This implies there will be more potential transit users in the future.
Senior citizens: Senior citizens are individuals between over the ages 65 to 79. This population is reported in two age cohorts: 65 to 79 and 80Plus. This sector of the population is also considered as one who uses transit as a mode of transportation . 20150 data shows that 12x percent of the population in the Flagler Corridor are included in this category. The gGrowth in elderly population is estimated to be slightly higher in the Ccorridor than in the County. The estimated numbers increase by by 15 percent in the County compared to 25% in the Corridor by 2040.
Low income households: Households with an annual income less than $25,000: In 2015, there were about 55,000 low-income households which is about 37 percent in the Corridor where the County consists about 30 percent low-income households. The Flagler Corridor is estimated to see retain a slightly higher growth share of low income households at 38 percent than the County at 30 percent by 2040 .
Plot of demographic trends
The following are the list of data files used for the analysis in this document.
DAILY_TEM4_300Iter.dbf : is the final ODME subarea loaded network. The initial settings of 12 loops showed the ODME process did not converge, where the %RMSE between the Count and Volume was checked. The convergence criteria was set at 15% for subarea wide RMSE. This condition was not met within default 12 iterations and thus additional iterations up to 300 feedback loops were run. The %RMSE at the end of 12 iteration was only 27% which clearly shows the ODME needs to be re-run. At the end of 300th iteration, the %RMSE is still at 17% and typically for a subarea wide it should be under 14%. Meaning more iterations can be run, however, before making that call, we need to understand the ODME results and why it didn’t converge.
before_ODME.dbf : is the processed subarea network links with volumes prior to ODME. The networks are time period specific and carry regional nodes numbering. The subarea network file with old and new nodes is used to convert the regional networks to one single dbf file with just volume and counts by time periods.
regional_before_ODME.dbf : is the final loaded regional assignment network. The links and nodes are subset to represent subarea region. Technically before_ODME.dbf and regional_before_ODME.dbf should be identical, however, they are not although very close. Most of the tables show both before ODME volumes.
DAILY_TEM4_Link.shp : final ODME results, same as DAILY_TEM4_300Iter.dbf but a shape file.
PB_Flagler_Count_Location.csv : List of subarea links where PB collected traffic counts.
PB_Flagler_Counts.csv : PB counts
Flagler_Subnet_Links.csv : All Flagler Corridor links where there are counts (not PB collected counts).
The latest delivered model from Feb 08, 2017 was run with 2010 seed skim with ODME turned on. The ODME procedure was applied by time period. For each of the time periods, root mean square error was generated by volume group, facility type and area type. However, the final validation statistics were computed on the daily results, which are not a true depiction of the ODME results. Thus additional statistics such as Root Mean Square Error (RMSE and %RMSE), Correlation Coefficient R2 were computed by time periods.
Additionally, stats from the before ODME procedure were looked to assess the ODME improvement. There are two way to get loaded network before ODME process.
The before ODME volumes were extracted from the final loaded network of the regional highway assignment **{OUTDIR}_LOADED.net** . However, the counts in this network seemed not updated and are different from the subarea network counts. For this reason, the ODME before and after comparisons with this network may not give correct assessment of the ODME improvements.
Before initiating the ODME procedure, the subarea OD trip tables were generated for each time period by assigning the regional OD trip tables to the regional network with subarea network option turned on. This resulted in a subarea OD matrix as well as a loaded regional network for each time period. The following loaded regional networks were exported to dbf links and post processed to retain volumes.
In order to compare the link volume from the regional model (before ODME) to subarea model (after ODME), the links nodes should be mapped. This mapping is done by exporting subarea.net to both link.dbf and node.dbf files, where node.dbf contains node correspondence between regional and subarea networks, and link.dbf contains the links that can be be compared to subarea ODME results.
ODME process in cube has many options to choose the targets from links counts, screen-lines, trip ends (zonal origins and destinations), paths, previously developed OD matrix and combination of these. The general method is the loaded network is converted into a specific format called ICM file, for efficient use where volumes, counts and other path choice settings are specified. Along with this file, screen-lines (link count locations) can be supplied and OD trip ends can be supplied. Each of these targets contains confidence thresholds that sets priority to targets. For this application the following targets and confidences were used:
High level overview of the process:
So, the only standard report from ODME process is the report produced in step 5 where the process lists ODME results for each ODME iteration. The convergence criteria used here are
Some of the stats: observed counts, model volumes, RMSE, %RMSE, R-Squared and Scattered Plots were computed by:
for the following time periods
Regional - Before ODME : The AM volumes before ODME process were extracted from the regional assignment. The total number of links with counts in the subarea are only 208. It is extremely hard to tell the year for this count although the field CNT_AM_10 might imply 2010 counts. The subarea wide %RMSE is 62% and the R2 is at 0.82. Most of the low volume and high volume groups shows R2 under 0.5
Subarea - Before ODME : Theoretically this should be same as above. Since the ODME subarea OD matrix extraction process uses a re-assignment of the regional trips tables, a new loaded highway networks are generated. The assignment settings here match to regional assignment. The model volume by volume group (volBin) shows significant difference from the above table. The count in this table are updated with ODME counts for the subarea. The subarea wide R2 is slightly better at 0.84, however it is hard to tell if this is due to change in counts or more number of counts (256 counts instead of regional 208 counts as above).
Subarea - After ODME : Shows ODME results, where R2 for subarea is at 0.96 and %RMSE is at 17%. The ODME feedback loop at 300th iteration shows %RMSE of 17.7% and ideally this should be under 14% meaning more number of iterations or more counts are required.
The plots below show a comparison between ODME counts and ODME results by volume group, facility type and area types.
Most of the data points by volume groups align well, however, there are few outlier in 1 - 5,000 group and 10,000 - 20,000 group.
Most of the data points fall close to the “best fit line”, however there are few that fall way off the line.
Most of the ODME results looks fine by Facility Type, however, types 1 and 7 show some link volumes way over or under. These links need to be studied further to determine if there is a count issue.
Most of the ODME results looks fine by Area Type, however, types 3 and 4 show some link volumes that are way over or under. These links need to be studied further to determine if there is a count issue.
Some of the stats: observed counts, model volumes, RMSE, %RMSE, R-Squared and Scattered Plots were computed by:
for the following time periods
Some of the stats: observed counts, model volumes, RMSE, %RMSE, R-Squared and Scattered Plots were computed by:
for the following time periods
The daily highway stats were computed for before and after ODME links.
There are a total of three sets of counts as described below:
PB collected counts and processed them for use in ODME, however,these counts weren’t used in the ODME process.
Table below shows all three above counts and three volumes: Regional volume, Subarea Volume before ODME and after ODME.
In order to better understand the ODME results, not just at locations where PB collected the data but all along the Flagler Corridor, all Flagler links with available / network coded counts were studied. Table Flagler Links with ODME Counts below shows the links, ODME count and ODME results along with regional volumes, counts and PB counts where available.
A total of 56 locations along Flagler Corridor contains counts used for ODME. Table Summary of Flagler Links with ODME Counts shows, out of 56 links with ODME counts along Flagler, about 33 are within 10% of the observed ODME count.
Of these 56 locations, PB collected data was available at 12 locations. Click the empty cell / box under “PB_count_locations” then move slider to 1 to show select all rows where PB counts are available. For the rows where the % percent difference is high, the ODME results match close to PB counts. This clearly demonstrates that if we used PB counts in addition to ODME counts, the ODME results would have been lot better.